Maximum Drawdown: Reading a Peak-to-Trough Decline

Two accounts finish the year at the same value, built from the same set of returns. One you could have held without much stress. The other spent months trading below its old high, and plenty of people would have sold near the low. What separates them is maximum drawdown, the deepest fall from a prior peak to a later low over a stated window. Two portfolios can post the same average return and the same volatility and still hand you very different rides, because a drawdown depends on the order losses arrive in, and that order is invisible in an average.

What maximum drawdown actually measures

Start with a series of values. It can be an account balance marked at each day’s close, or the adjusted price of a single stock. Keep a running peak, the highest value the series has reached up to that point. At every observation, measure how far the current value sits below that running peak, in percent. The drawdown at any moment is that gap. The maximum drawdown is the deepest gap the series ever printed over the window you’re studying.

Here’s a short sequence I use when I teach this. The account moves 100, 110, 105, 120, 96, 108, 132, 118, 140. Track the running peak beside it: 100, 110, 110, 120, 120, 120, 132, 132, 140. Now read the gaps. At 105 the account is 4.5% under its 110 peak. At 96 it’s 20% under the 120 peak, and that’s the deepest reading in the whole path. That 20% is the maximum drawdown.

The first number I point at is the 96, but the drawdown clock actually starts at the 120 peak that came before it. Depth is measured from the high-water mark, so the peak is where the episode begins and the low is only where it bottoms.

Four dates hide inside one number

A single figure like “-20%” collapses a whole story into one reading, and that story has at least four markers worth pulling apart. There’s the start date, which is the prior peak. There’s the trough date, where the low prints. There’s the recovery date, the first time the series climbs back above the old peak. And there’s the recovery duration, the stretch of time from peak to recovery.

In the sequence above, the peak sits at 120, the trough at 96, and recovery arrives at 132, the first value that clears the old 120 high. The account stayed below its high-water mark for three observations before it recovered. The 20% depth tells you nothing about that gap. A -20% drawdown that heals in three weeks and a -20% drawdown that grinds on for three years read as the same number, and they’re nowhere near the same problem for the person holding the position.

I keep the Ulcer Index next to maximum drawdown in my own notes for exactly this reason. Depth answers how far, and a duration-sensitive measure answers how long, which is the half of the experience a single worst-case figure quietly drops.

Why the order of losses changes everything

This is the part that surprises people. Take four period returns: down 10%, down 10%, up 12%, up 12%. Multiply them in any order and the account lands in the same place, near 101.6 from a start of 100. The average return is identical. The standard deviation of those returns is identical too, since it’s the same four numbers. Only the sequence changes.

Put the two losses first: 100 falls to 90, then to 81, then recovers to 90.72 and finally 101.6. The running peak stayed at 100 until that last step, so the deepest gap sits at 81, a maximum drawdown of 19%. Now spread the losses out: 100 rises to 112, dips to 100.8, climbs to 112.9, then eases back to 101.6. The deepest gap this time is only 10%, measured off the 112 and 112.9 peaks.

Same returns, same average, same volatility, same ending value, and the maximum drawdown nearly doubles from 10% to 19% purely because the losses clustered. That’s why an equity curve that looks calm on summary statistics can still hide a punishing worst stretch. Volatility treats an up move and a down move as the same distance from the mean. Drawdown weighs only how far you sit below the last high, so back-to-back losses bite it in a way they never bite a volatility number.

The math of getting back to even

Recovery is its own axis, and the arithmetic runs against you. A 20% drawdown takes a 25% gain to repair, because you’re now growing a smaller base back up to the old peak. A 50% drawdown takes a 100% gain to get back to even. A 10% dip takes 11.1%. The deeper the hole, the more lopsided the climb out.

Because bet size drives how deep those holes get, the mathematics of sizing and drawdown is a field of its own. Ralph Vince‘s work on optimal position sizing sits right at that intersection, mapping how a bigger bet buys a higher expected return at the cost of a steeper worst-case decline.

Time to recover is a separate question again, and it hangs entirely on how strong the returns are after the trough. A shallow 8% decline can sit unrepaired for a year if the market only drifts. A brutal 30% decline can heal in a couple of months inside a sharp rebound. Depth and recovery time move independently, so you can’t read one from the other. That’s why quoting a maximum drawdown without its recovery duration leaves half the picture out.

A drawdown does not need a market crash

People often assume a large drawdown means the whole market fell apart. It doesn’t have to. A single stock can carve a 35% peak-to-trough decline while the broad index sits within a few percent of its own high, because company-specific news moves one name and leaves the rest alone. A long-only strategy can do the same when its style falls out of favor.

This is where the difference between an absolute decline and a benchmark-relative decline matters. Absolute drawdown asks how far your capital fell from its own peak. Relative drawdown asks how far you slipped behind a benchmark. They answer different questions, and a position can look fine on one while bleeding on the other. A fund that drops 5% in a year the index gained 20% never posted a frightening absolute drawdown, yet it fell 25% behind on a relative basis. Decide which question you’re asking before you quote a single figure.

Diversification works on the same lever. Spreading capital across positions that don’t fall together tends to shrink the deepest drawdown, because a loss in one holding often lands on a day another is flat or rising, and the running peak barely dents. The catch shows up in a panic. Correlations jump toward one, everything sells off at once, and that cushion thins out exactly when you were counting on it, which is why the worst drawdowns tend to arrive when diversification was supposed to help most.

What traders build on it, and where it misleads

Once you can measure the worst stretch, you can act on it. Some traders set hard drawdown stops, cutting size or standing aside once an account falls a fixed percentage from its peak. Position sizing leans on the same idea: keep the plausible drawdown small enough that a bad run can’t end the account. That question, the odds of losing so much you can’t continue, is the domain of risk of ruin.

The trap is treating a historical maximum drawdown as a ceiling. It’s one realized episode from one sample, and the next one can run deeper. A strategy whose worst drawdown was 18% across a five-year backtest won’t politely stop at 18% in live trading. Skewness and fat tails make the rare, larger loss more likely than a tidy history suggests, and any measure built from past downside episodes, whether a drawdown figure or a conditional value-at-risk number, describes what already happened. Use the figure to size and to prepare, and hold it loosely as a guide rather than a promise.

Researchers who study momentum sometimes look at the drawdown path itself, beyond the final return, because how a strategy recovers after a deep decline carries information about how it may hold up the next time. That’s one reason drawdown and recovery turn up alongside returns in trading research rather than as an afterthought.

Reading one episode honestly

Every maximum drawdown you’ll ever quote is shaped by choices you might not notice. Measure it on daily closes and you catch the swings between one close and the next. Measure it on month-end values and you smooth the same period into something milder. The sample window, the data frequency, the valuation method, and the starting point all move the number. Change the window and the worst episode can change with it.

So treat the figure for what it is: an honest label on the single worst peak-to-trough decline that actually happened, over the exact period you measured. It reports the depth of one realized loss. It can’t tell you the odds of the next one, what will cause it, or when it will land. Read it with its trough date and its recovery time attached, and it becomes a genuinely useful description of how an equity curve behaved. Read it as a forecast and it will quietly mislead you.

Learn the pattern. Ride the trend. Keep the gains.

Educational content only. Not investment advice. Trading involves risk. You are responsible for your decisions.